<?xml version="1.0" encoding="utf-8" standalone="yes"?>
<rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom">
  <channel>
    <title>元认知 on AI内参</title>
    <link>https://www.neican.ai/tags/%E5%85%83%E8%AE%A4%E7%9F%A5/</link>
    <description>Recent content in 元认知 on AI内参</description>
    <generator>Hugo</generator>
    <language>zh-cn</language>
    <lastBuildDate>Mon, 06 Jul 2026 10:40:07 +0800</lastBuildDate>
    <atom:link href="https://www.neican.ai/tags/%E5%85%83%E8%AE%A4%E7%9F%A5/index.xml" rel="self" type="application/rss+xml" />
    <item>
      <title>止损的艺术：CONVOLVE 揭示 AI Agent 从“盲目执行”向“理性认知”的进化拐点</title>
      <link>https://www.neican.ai/insights/convolve-ai-agent--20260706104007260-0/</link>
      <pubDate>Mon, 06 Jul 2026 10:40:07 +0800</pubDate>
      <guid>https://www.neican.ai/insights/convolve-ai-agent--20260706104007260-0/</guid>
      <description>华盛顿大学开发的 CONVOLVE 框架通过上下文工程赋予 AI Agent“及时止损”的决策能力，这一研究不仅破解了 Agent 在复杂任务中盲目执行的难题，更通过经验迁移机制展示了企业降低 AI 部署成本、提升自主系统安全性的商业前景。</description>
    </item>
    <item>
      <title>AI元认知萌芽：Anthropic「概念注入」实验揭示心智黑箱与信任鸿沟</title>
      <link>https://www.neican.ai/insights/aianthropic-20251030184007815-0/</link>
      <pubDate>Thu, 30 Oct 2025 18:40:07 +0800</pubDate>
      <guid>https://www.neican.ai/insights/aianthropic-20251030184007815-0/</guid>
      <description>Anthropic的“概念注入”实验首次证实，大型语言模型（LLM）已具备初步的内部自我审查和认知控制能力，颠覆了传统认知。这项技术突破为理解AI“心智”黑箱提供了前所未有的显微镜，揭示模型可在输出前察觉内部“思想”并区分“意图”。然而，模型内省能力的不稳定性与自我报告的不忠实性，正加速AI可解释性成为构建可信AI系统、应对潜在“隐藏意图”及重塑人机信任关系的关键产业与伦理挑战。</description>
    </item>
    <item>
      <title>大模型的“思维盲区”：DeepMind揭示推理致命弱点，颠覆AI安全与信任边界</title>
      <link>https://www.neican.ai/insights/deepmindai-20250703154004154-0/</link>
      <pubDate>Thu, 03 Jul 2025 15:40:04 +0800</pubDate>
      <guid>https://www.neican.ai/insights/deepmindai-20250703154004154-0/</guid>
      <description>DeepMind的最新研究揭示，大模型在推理过程中对自身错误与无关信息缺乏“元认知”能力，且越大模型越难以自愈，甚至更易受新型“思考注入”攻击影响。这颠覆了“大模型更安全”的传统观念，对AI的可靠性、商业部署及社会信任构成严峻挑战，促使业界深思如何赋予AI真正的自省与纠错能力。</description>
    </item>
  </channel>
</rss>
